Accurate estimates of snow water equivalent (SWE) are essential for understanding hydrological processes and managing effective management of water resources, particularly in snow-dominated regions. Many methods for estimating SWE rely on in-situ measurements and/or numerical models. In-situ measurements, such as those provided by the USDA Snotel network, have the advantage of being direct observations of SWE but are only sparsely available and suffer from challenges of representativity. At the same time, numerical models embed knowledge of the physical processes underlying the snowpack accumulation and ablation but can be computationally expensive to run over large areas. In this study, we investigate applying deep learning techniques to predict the spatiotemporal distribution of SWE from a combination of atmospheric forcings derived from the Weather Research and Forecasting (WRF) model, geographic parameters related to topography and land cover that influence snow persistence, and historical observations of snow presence/absence from remote sensing data. By leveraging static variables and dynamic atmospheric forcings from WRF  as input features, we train a convolutional long short-term memory (ConvLSTM) network to predict SWE. Our proposed deep learning model aims to accelerate the prediction of spatially distributed SWE compared to traditional methods and can complement process-based land surface models often used to predict SWE. The computational savings associated with training and forward integration of machine learning based models open the door to high-resolution ensemble forecasting of SWE and assimilation of observations for real-time SWE estimation.

Ecosystem dynamic models have been widely used to estimate terrestrial carbon flux and to project ecosystem structure and composition over time and space, because of their efficiency over direct field measurements and easy applicability to broader spatial coverage. However, such models have also been associated with internal uncertainties, as well as complexities arising from distinct qualities of the ecosystem being analyzed. The widespread sagebrush-steppe ecosystem (dominated by Artemisia spp.) in Western North America holds high ecological and social significance, but is threatened by anthropogenic forcing factors, including impacts from invasive species, climate change, and altered fire regimes. To restore the ecosystem, land managers have focused on reducing flammable vegetation and seeding native species. However, the collective effects of restoration activities, fire, climate change, and invasive species on ecosystem dynamics are poorly understood. We applied the Ecosystem Demography (ED2) model to analyze its effectiveness in predicting plant function type (PFT) composition and ecosystem fluxes, parameterized and validated using empirical datasets for different carbon, vegetation and fire scenarios at Reynolds Creek Experimental Watershed (RCEW), Idaho, USA. We initialized ED2 with 20 x 40 grids of 1 km resolution representing and allowed PFTs to grow for 20 years to reach an equilibrium state. Results showed shrubs dominating C3 grass in a few years of time, sooner for increased CO2 and initial ecosystem condition. A separate scenario with potential fire showed significant loss in biomass within eight years of time. Results from this modeling study can improve our understanding of broad-scale ecosystem processes in sagebrush-steppe landscapes and inform land management and restoration strategies.

Rooting depth is an ecosystem trait that determines the extent of soil development and carbon cycling. Recent hypotheses propose that human-induced changes to Earth’s biogeochemical cycles propagate deeply due to rooting depth changes from agricultural and climate-induced land cover changes. Yet, the lack of a global-scale quantification of rooting depth responses to human activity limits knowledge of hydrosphere-atmosphere-lithosphere feedbacks in the Anthropocene. Here we use land cover datasets to demonstrate that global rooting depths have become shallower in the Anthropocene, and are likely to become yet shallower this century. Specifically, globally averaged depths above which 99% of root biomass occurs (D99) are 8.7%, or 16 cm, shallower relative to those for potential vegetation. This net shallowing results from agricultural expansion truncating D99 by 82 cm, and woody encroachment linked to anthropogenic climate change extending D99 by 65 cm. Projected land cover scenarios in 2100 suggest further D99 shallowing of 63 to 72 cm, exceeding that experienced to date and suggesting that the pace of root shallowing will quicken in the coming century. Losses of Earth’s deepest roots—soil-forming agents—suggest unanticipated changes in fluxes of water, solutes, and carbon. Our work constrains rooting depth distributions for global models, allowing the land modeling community to explore cascading effects of rooting depth changes on water, carbon, and energy dynamics, and can guide design of field-based efforts to quantify deep anthropogenic influences. Understanding human influence on biota’s reach into Earth’s subsurface will improve predictions of interactive functioning of the biosphere, lithosphere, and hydrosphere.

Rooting depth is an ecosystem trait that determines the extent of soil development and carbon (C) and water cycling. Recent hypotheses propose that human-induced changes to Earth’s biogeochemical cycles propagate deeply due to rooting depth changes from agricultural and climate-induced land cover changes. Yet, the lack of a global-scale quantification of rooting depth responses to human activity limits knowledge of hydrosphere-atmosphere-lithosphere feedbacks in the Anthropocene. Here we use land cover datasets to demonstrate that root depth distributions are changing globally as a consequence of agricultural expansion truncating depths above which 99% of root biomass occurs (D99) by ~60 cm, and woody encroachment linked to anthropogenic climate change extending D99 in other regions by ~38 cm. The net result of these two opposing drivers is a global reduction of D99 by 5%, or ~8 cm, representing a loss of ~11,600 km3 of rooted volume. Projected land cover scenarios in 2100 suggest additional future D99 shallowing of up to 30 cm, generating further losses of rooted volume of ~43,500 km3, values exceeding root losses experienced to date and suggesting that the pace of root shallowing will quicken in the coming century. Losses of Earth’s deepest roots — soil-forming agents — suggest unanticipated changes in fluxes of water, solutes, and C. Two important messages emerge from our analyses: dynamic, human-modified root distributions should be incorporated into earth systems models, and a significant gap in deep root research inhibits accurate projections of future root distributions and their biogeochemical consequences.